Eventos Anais de eventos
COBEM 2021
26th International Congress of Mechanical Engineering
A Machine Learning Approach to Solve the Lid Driven Cavity Flow
Submission Author:
Rogerio Werneck Costa Rodrigues Filho , DF
Co-Authors:
Rogerio Werneck Costa Rodrigues Filho, Adriano Rosa
Presenter: Rogerio Werneck Costa Rodrigues Filho
doi://10.26678/ABCM.COBEM2021.COB2021-0667
Abstract
The growing production of data has made possible to use machine learning algorithms on mechanical engineering problems that were not previously viable due the lack of data. Traditional numerical simulation sometimes might have a huge computational cost, in such way that real time simulation might not be possible. Some numerical simulation softwares has started to allow its users to use computational clouds to perform a faster simulation,in such a way that these companies have started to develop an abundant data base with computer aided engineering (CAE) problems. Also, highly monitorated locations usually have a lot of sensors creating a log of historical data, for example temperature inside a data center along time, in a such a way that modeling the physical conditions inside this site is more viable using machine learning techniques than traditional numerical simulation. Since machine learnign algorithms do not require mathematical models of the physical problem to get real results, the use of these techniques would facilitate the entire process of modelling the problem when comparing to traditional simulation. In this work, a machine learning model focused on lid driven cavity flow, is developed using feedforward neural networks besed on data results obtained by traditional numerical methods, such as second order projection method. The use of these techniques does not require the use of Navier-Stokes equation to create reliable results. This means that, the algorithm is able to create its own mathematical model based only on data. Also, machine learning algorithms, after the training process, do not require a lot of computational power to process data, in such a way that the simulation time has been drastically reduced when compared to conventional numerical simulation methods. The final results precision, using machine learning techniques, has not been significantly affected in a way that is made possible to obtain reliable results through this method.
Keywords
machine learning, 2D lid-driven cavity flow, Artificial neural networks, CFD
