In structural mechanics, Finite Element Method (FEM) is the ruling numerical simulation method. FEM has limitations that, it is numerically expensive for linear as well as non-linear large size problems. If the input parameters in FEM need to be adjusted, even slightly, simulation must be re-run to check the sensitivity of changed parameter. To reduce computational time in these cases, machine learning (ML) and neural networks (NN) surrogate models are proposed to solve linear and non-linear structural mechanics problems with a change in loading conditions for the same geometry. The applied load on the same model was considered as input, and displacements at the nodes were considered as output. The ML model was developed using the sklearn library and an NN model by using the keras library in Python. Root means square error (RMSE) and accuracy in percentage these performance matrices were used to compare ML and NN models with FEM. Gaussian Process Regressive (GPR) was most reliable model in case of linear cantilever beam and 3D bolt which gives highest percentage accuracy between AI and FEM for testing dataset of 99.99% and 99.81% respectively, whereas for non-linear cantilever beam Decision Tree (DT) was the reliable model with percentage accuracy of 99.87%.