Predicting the simulation result of locomotive components using ANN

Abstract

This thesis investigates the application of a physics-informed artificial intelligence (ai) model for the structural analysis of locomotive components, with a specific focus on the blower cab, a critical subsystem responsible for ventilation in railway locomotives. The research investigates the ability of the AI model to forecast the structural behaviour of the blower cab under different pillar thicknesses, utilizing a dataset of simulation files to train, evaluate, and predict results. The methodology encompasses the generation of blower cab models with thicknesses ranging from 3 mm to 6.5 mm, finite element analysis (FEA) to produce result files, and the training of a physics-informed neural network to predict displacements. Testing on six configurations (3.5 mm and 5.5 mm thicknesses) yields accuracies exceeding 97%, with mean absolute errors (MAE) of 0.705 mm and 0.239 mm, respectively, and prediction times of less than 1 minute compared to 20 minutes for FEA. Predictive results for thicknesses of 4.5 mm and 6.5 mm (22.36 mm and 22.25 mm maximum displacements) align with structural mechanics principles, demonstrating a consistent decrease in displacement with increasing thickness. The model's high accuracy can be attributed to minimal design variations, although slight reductions in accuracy for specific configurations indicate sensitivity to the range of training data.