Predicting Transient Temperatures during Heat Conduction using Ensemble Regression Models
The training dataset for 1D, 2D and 3D Heat conduction has been generated by virtue of simulations and stored as spreadsheets. These training masterdatasets are used to train and develop the Multi Linear Regression and Random Forest Models for the three cases respectively. The R2 scores of Optimized RF regressors are found to be ~98%. One can use the dataset to train a new model as per requirement.
The materials used for simulation are: Copper (Diffusion Coefficient = 1.11E-04) Aluminium 6061 (Diffusion Coefficient = 6.4E-05) iron (Diffusion Coefficient = 2.3E-05) Steel Stainless 310 (Diffusion Coewfficient = 3.55E-06)
Other materials can be used to generate more training data.
For this project simple geometries have been considered i.e., Rod (Straight line), Plate (Rectangular) and Block (Cuboidal). Other more complex shapes can be used to model a larger and more inclusive dataset.
A generalized 3D Predictor program has also been built which can be used directly to predict Temperature values at an instant or values corresponding to an entire dataset. This takes a runtime of few milliseconds as against tens of minutes or sometimes hours of numerical simulations and physical experimentations.
The intention of the work is to create an open source environment with relevant datasets and accurate ML based models for researchers in the domain to further predict the Temperature values in an instant.
Thanks for reading and Goodluck with your research!