Deep Learning for Engineering Problems

In this paper, we propose a data driven deep learning model to solve transport phenomenon without incorporating physics based PDEs. Here, a deep Recurrent Neural Network (RNN) with Long Short Term Memory (LSTM) is used to solve a general 2D heat conduction phenomenon. The problem is solved for square and circular geometries. The model is trained on 1670 cases for square domain and 3000 cases for circular domain. The cases include both Dirichlet boundary conditions and Neumann boundary conditions. The model was then tested on 610 cases for square geometry and 1000 cases for circular geometry. Our proposed RNN-LSTM model shows a 3-order speed up in computational time compared to conventional finite difference method. Moreover, the predicted solution shows 99.9% accuracy. Also, our proposed model can easily be generalized and extended for other transport phenomenon problems, both linear and nonlinear. We test this by considering a simple nonlinear advection-conduction phenomenon. We believe the RNN-LSTM deep learning method has the ability to predict transport phenomenon in applications like aerospace, automobile, semiconductor and thermal management domains like electronic cooling applications, without incorporating physics based PDEs.