Course Projects
Artificial neural networks for predicting the performance of membrane distillation
October - December 2020
Report, 8-minute presentation
- Machine learning and artificial intelligence can be applied in desalination to capture the complex heat, mass transfer physics and predict their system scale performance. In this project, we trained artificial neural networks (ANNs) on an extensive dataset consisting of the permeate flux and energy efficiency of MD systems as a function of feed salinity, temperature, mass flow rate and module length.
- Hyperparameter optimization was carried out to determine the optimal number of hidden layers, learning rate, activation function and SGD batch size for the ANN eventually leading to an accuracy of 0.94.
- In conjunction, a simple multivariable regression model was developed to compare the results from ANN, regression and multiphysics computations on unseen operating conditions.
- This project marked my first implementation of AI tools to solve practical problems and was completed using the concepts from an introductory course on machine learning (ME597).
Method of characteristics to evaluate steady flow in a supersonic nozzle
March - May 2020
Report
- Formulated a generalized code to resolve the two-dimensional steady flow inside a supersonic nozzle using the method of characteristics for different number of expansion waves.
- Designed nozzle contours to achieve Mach 3.0 at the exit using 4, 5, 7 and 9 expansion wave test cases and determined the minimum nozzle length required to achieve this condition.
- Established convergence of the solution with number of expansion waves and compared the solution with results from quasi one-dimensional analysis.
Unsteady flow past a bluff body
October - December 2019
Report, Presentation
- Developed a finite difference algorithm to resolve the unsteady flow past a square cylinder using the immersed boundary method for Reynolds number ranging from 10-1,000.
- Implemented ghost noding for the pressure equation at the boundaries to reduce the computational effort and condition the resulting coefficient matrix.
- Validated results for Stokes flow and low Reynolds number regimes with compact scheme computations. Please refer to the report for in-depth discussions on numerical schemes and results.
Constrained optimization using conjugate gradient method
March - May 2019
Report
- Formulated a Fortran code for multi-variable constrained optimization using the conjugate gradient method. A comparison was drawn between the usage of bracket-operator penalty method and the method of multipliers to handle the problem constraints.
- Introduced a reset option to facilitate convergence and the unidirectional search was carried out using a combination of bisection and bounding phase method.
Two dimensional unsteady laminar convection in a confined channel
March - May 2019
Report
- Implemented a streamfunction-vorticity formulation using the finite difference method to solve Poiseuille Benard problem in a confined channel.
- Formulated an algorithm to solve the linear system using BiCGSTAB (biconjugate gradient stabilized) method with preconditioning from SIP.
- Qualitatively studied the effects of open boundary conditions on vorticity and temperature contours.
Design and Development of a Formula SAE Car
July 2017 - February 2018
- Designed and manufactured an efficient aerodynamic package consisting of front wing, sidepods and diffuser for a Formula SAE car using ANSYS Fluent.
- Modeled the interaction of nose cone contour and front wing design to optimize the aerodynamic performance and effectively channel airflow to the sidepods.
- Successful in achieving an additional downforce of 250 Newtons with a resulting drag coefficient of 0.14 from the package.