Akolekar, Harshal

Biomimicry involves drawing inspiration from nature’s designs to create efficient systems. For instance, the unique herringbone riblet pattern found in bird feathers has proven effective in minimizing drag. While attempts have been made to replicate this pattern on structures like plates and aerofoils, there has been a lack of comprehensive optimization of their overall design and of their constituent individual repeating structures. This study attempts to enhance the performance of individual components within the herringbone riblet pattern by leveraging computational fluid dynamics (CFD) and supervised machine learning to reduce drag. The paper outlines a systematic process involving the creation of 107 designs, parameterization, feature selection, generating targets using CFD simulations, and employing regression algorithms. From CFD calculations, the drag coefficients (C d ) for these designs are found, which serve as an input to train supervised learning models. Using the trained transformed target regressor model as a substitute to CFD, C d values for 10,000 more randomly generated herringbone riblet designs are predicted. The design with the lowest predicted C d is the optimized design. Notably, the regressed model exhibited an average prediction error rate of 6% on the testing data. The prediction of C d for the optimized design demonstrated an error of 4% compared to its actual C d value calculated through CFD. The study also delves into the mechanics of drag reduction in herringbone riblet structures. The resulting optimized microstructure design holds the potential for reducing drag in various applications such as aerospace, automotive, and marine crafts by integrating it onto their surfaces. This innovative approach could significantly transform drag reduction and open pathways to more efficient transportation systems.

The laminar kinetic energy (LKE) transition model recently has shown good predictions on separation-induced transition that frequently experienced by low pressure turbines (LPT). In contrast to LPTs, high-pressure turbines (HPT) often are subject to bypass transition which currently is not captured well with the LKE model. Because these two types are common transition modes for turbines, the current effort is focused on generalizing and improving the LKE model to be able to predict the different types of transition. In addition to modify the transition model to achieve better on-blade predictions, the turbulence model in the wake region is also revised for more accurately capturing the wake mixing. Hence, the main purpose of this study is to use a data-driven approach to simultaneously develop spatially separate transition and turbulence closures suitable for a range of different turbine configurations. To achieve this, two strategies are adopted. The first is to employ a multi-case multi-objective computational fluid dynamics (CFD) -driven model training framework. The training is performed on several LPT and HPT configurations, specifically the T108, T106A, and LS89 sections, that feature both separation-induced and bypass transition to ensure better generalizability of the models. The second strategy employed is the use of a newly derived set of local non-dimensionalized variables, that serves as the inputs for the LKE model corrections. Among the training cases, steady calculations are conducted for LPTs. The LS89 case is an HPT characterized by shocks, acoustic waves, and unsteady trailing edge vortex shedding. To capture these, for the first time an unsteady solver is utilized during the CFD-driven training, and the time-averaged results are used to calculate the cost function as part of the model development process. The model obtained from the new training process are tested on the a steady case - T108 with a higher Reynolds number and an unsteady case - PakB profile. Their performance are assessed in terms of pressure coefficient, wall shear stress and wake losses.

In low pressure turbines (LPT), due to the low Reynolds number a large part of the blade boundary layer remains laminar and transition may occur due to flow separation. The boundary layer details at the blade trailing edge can change substantially depending on the transition region topology and can strongly influence the wake mixing occurring downstream. Accurately predicting these flow phenomena still poses a challenge for Reynolds averaged Navier-Stokes (RANS) and unsteady RANS methods. In this work a recently developed computational fluid dynamics (CFD) driven machine learning framework featuring multi-expression, multi-objective optimization is exploited for the first time to simultaneously develop transition models and turbulence closures in a fully coupled way, aimed at improving both transition and wake mixing predictions in LPTs. The T106A blade cascade with an isentropic Reynolds number of 100,000 is adopted as a training case. The baseline transition model is based on a laminar kinetic energy transport approach, and the machine learning approach is used to reformulate the source terms as functions of suitably defined non-dimensional ratios. Additionally, machine learning based explicit algebraic Reynolds stress models are used to improve wake mixing predictions, making use of a specifically and newly developed wake sensing function based strategy that allows an automated zonal application of the developed models. It is shown that both on-blade performance and wake mixing can be predicted accurately with data-driven transition and turbulence models that have benefited from CFD feedback in their development, ensuring that their mutual interactions are captured.

Large language models (LLMs) such as ChatGPT and Google Bard have garnered significant attention in the academic community. Previous research has evaluated these LLMs for various applications such as generating programming exercises and solutions. However, these evaluations have predominantly been conducted by instructors and researchers, not considering the actual usage of LLMs by students. This study adopts a student-first approach to comprehensively understand how undergraduate computer science students utilize ChatGPT, a popular LLM, released by OpenAI. We employ a combination of student surveys and interviews to obtain valuable insights into the benefits, challenges, and suggested improvements related to ChatGPT. Our findings suggest that a majority of students (over 57%) have a convincingly positive outlook towards adopting ChatGPT as an aid in coursework-related tasks. However, our research also highlights various challenges that must be resolved for long-term acceptance of ChatGPT amongst students. The findings from this investigation have broader implications and may be applicable to other LLMs and their role in computing education.