Embedded Files
INDUSTRY PROJECTS
Project Summary
Imagine if user can set up simulations by typing his request using natural language!
In this project, I combine Large Language Model (LLM) with AI-driven simulation to make simulation setup an intuitive process and get quick simulation results.
This project is a teaser of how the future of simulation will be.
Check out Ansys AI LinkedIn page.
Project Summary
imPULSE mixing tanks play a pivotal role in the pharmaceutical industry, serving as essential vessels for the blending of various components in the production of medicines and pharmaceutical products. Precise control of parameters like agitation speed, temperature and pressure is crucial to maintaining the integrity of sensitive compounds. Deep understanding of fluid dynamics would contribute significantly to optimizing the design and efficiency of these crucial mixing tanks, enhancing their performance and reliability in the pharmaceutical manufacturing process.
Description
Mixing in imPULSE mixers is achieved by the reciprocating motion of the plunger inducing fluid circulation inside the tank. Additionally, the plunger has flaps that open in the upstroke and close in the downstroke. In theory, the flaps motion can be modeled as a fluid-structure interaction (FSI) problem, but this will be computationally expensive. Instead, to achieve an efficient model, flaps motion were modeled virtually as porous media zones with variable resistance, such that it has a very high flow resistance in the downstroke corresponding to closed flaps, and has a very low resistance in the upstroke corresponding to opened flaps. This virtual representation was achieved by using a parametrized user-defined function (UDF) complied into Ansys Fluent software. The reciprocating motion of the plunger was modeled using the 'Layering' dynamic mesh approach, which is the most efficient dynamic mesh approach.
The simulation enabled engineers to estimate the blend time and the power consumption, with minimal computational time and hardware requirement.
Results
This work was a collaborative work within Ansys Fluids team, and was presented at AIChE Annual Meeting 2023 Conference.
Work Featured on Ansys Fluids LinkedIn page.
Project Summary
Developing customized Web Apps on top of reduced order models (ROMs) empowers non-specialists and less experienced engineers to utilize simulations efficiently and get insights into the complex dynamics of their industrial processes without the need of learning complex simulation software.
In this project, I develop a 3D field reduced order model (ROM) for temperature distribution prediction as a function of design inputs. This offers an easy-to-use, interactive web app which provides instantaneous visualization of 3D simulation predictions at early design stage, accelerating the design process.
Project Summary
Read the details of this project in this ANSYS Blog: Discover Machine Learning Insights for Physics-accelerated Mixing Process Design.
System-level model integrating Reduced Order Model (ROM) and data fusion FMU for mixing tank application.
Project Summary
Concept of data fusion.
In several situations, physics-based models fail to fully resolve the physical effects, ending up by the model prediction deviating from experimental data and real-life behavior. Residual modeling is a machine learning approach that aims to learn the residual error between the model predictions and the real-life measurements in order to correct the model predictions. The aim of data fusion is create an accurate predictive model by fusing few experimental data with several low-cost simulation data.
Description
Gearbox under consideration.
Engineers aim to quantify the power loss in gearboxes to optimize their designs and enhance their power transmission efficiency. They often rely on CFD simulations reduce the design costs and reach faster to the market. There are several CFD methods for modeling gearboxes which differs in computational time and accuracy. In theory, the moving and deforming mesh (MDM) method is an accurate approach; however, it has several challenges regarding the geometrical properties of the volume domain and its variations during the meshing cycle and the complexity to handle the mesh, which in several cases can result in an instable solution. On the other hand, the sliding mesh approach is an easy, stable, and computationally efficient approach as it relaxes the time step constraint and the mesh handling challenges; however, it has a lower accuracy.
In this project, I apply the Fusion Model within Ansys Twin Deployer tool to enhance the accuracy of the sliding mesh approach in predicating the power loss in gearboxes by fusing experimental data with CFD simulations data results.
Results
Results of data fusion modeling using Ansys Twin Deployer.
A 10X reduction in the design process were demonstrated, and the predicted power loss values from simplified sliding mesh approach were matching the experimental measurement upon utilizing the Fusion Model.
Page updated
Google Sites
Report abuse