21st Annual PKP Student Research and Fine Arts Conference: Oral Symposia VI
Using Machine Learning to Predict the Critical Reynolds Number of a Tandem Cylinder SystemFluid flow past two cylinders placed next to each other (tandem configuration) is common in engineering applications. We utilize machine learning techniques coupled with computational fluid dynamics simulation to predict the critical Reynolds number of a tandem configuration. First, we compute the flow behavior using Gerris. We find that for certain special choices of cylinder separation to diameter ratio the flow evolves from laminar to oscillatory (turbulent) behavior at a specific critical Reynolds number. While it is possible to extract the pressure data to analyze the transition, a computational bottleneck is the sheer volume of the generated information. In our work, we have utilized singular value decomposition (SVD) and principal component analysis (PCA) to identify the critical Reynolds number. These techniques allowed us to remove irrelevant simulation information by reducing the data matrix dimension. Based on our calculations we showed that it is possible to reconstruct the pressure around the cylinder using a minimal amount of data to predict the correct critical number.
Minimum flow rate in electro-coflowControlled generation of micron and sub-micron sized drops continues to be of strong interest for the scientific community due to the variety of applications in many different fields. Emulsion drops can be generated by flowing two immiscible liquids inside a glass-based microfluidic device. Their minimum size will be of the order of the tip size. To create smaller drops, an external electric field can be used, similarly to what it is done in the classical electrospray. In electrospray, a liquid is issued into air from an electrified needle. When the flow rate of the liquid is controlled, there is a minimum flow rate below which a cone-jet cannot be formed regardless of the applied voltage. This minimum flow rate gives you the minimum drop size that can be generated, usually one or two orders of magnitude smaller than the tip size. We explore this lower limit in electro-coflow using pressure control instead, and we have found a different result than in electrospray, with a more complex behavior. The use of pressure control and the presence of an outer moving fluid, enrich the dynamics in the minimum flow rate limit.
Feasibility of Reusable Radiochromic Plastics as DosimetersNew developments in the field of radiotherapy have created extremely effective and efficient procedures for the treatment of tumors. Such developments require complex radiotherapy systems and plans which have ultimately improved the successfulness of treatments and the options available to patients. However, these advanced treatments present challenges for current dosimetric verification techniques which struggle to keep up. We will be examining radiochromic plastics as a way of addressing this conflict. Radiochromic plastics are synthetic materials whose optical properties change upon absorption of dose. The color darkens when dose is absorbed, and the change in optical density is proportional to the dose absorbed. We examine a formulation which is designed to clear its response slowly after irradiation to determine if the formulation is reusable after it clears. If upon reirradiation the dose response remains linear, then reusability is an option. Our results show that the response remains linear over a range of five irradiations and a timespan of two years, but sensitivity drops around 20% after the first reirradiation and less than that upon further irradiation. The plastics clear at an exponential rate, but the time it took to clear increased after each reirradiation. These results suggest that a long-term reusable dosimeter is possible. Data on a faster clearing formulation will also be presented.
Data Driven Machine Learning Discovery of Fundamental Physical LawsMachine learning, which is part of artificial intelligence, has become an invaluable tool to manipulate, analyze, predict, and reveal trends and associations hidden within big data. Machine learning algorithms build a mathematical model of sample data in order to make predictions or decisions, whether simply filtering emails and recommending products in a search bar or discovering the fundamental laws governing highly sensitive chaotic systems. In this research investigation we apply the "Least Absolute Shrinkage and Selection Operator" (LASSO) method of data analysis that determines the relationship, or lack thereof, between variables, allowing for the removal of irrelevant features. The method is first applied to a generic system of differential equations, to demonstrate its applicability, before showcasing its application within the context of a chaotic Lorenz oscillator system. The generic coupled system is solved using the LASSO module available in Python's sci-kit-learn. A similar computational approach for the chaotic system with synthetic Gaussian noisy data successfully reproduces the original Lorenz attractor solution.