ANN MODEL

The Artificial Neural Network model was constructed to act as a surrogate for FEA. Key features include:

• Input Layer: Load, Design ID, Mass

• Hidden Layer: 10 neurons with tansig activation

• Output Layer: Predicts deformation, stress, stiffness

• Training Algorithm: Levenberg–Marquardt (LM)

• Dataset: 70% training, 15% validation, 15% testing

The ANN rapidly converged with low MSE (<10⁻⁴), indicating strong learning capabilities. It drastically reduces computational time, predicting results in under a second compared to lengthy FEA runs.

ANN PREDICTIONS & VALIDATION

Validation results demonstrate excellent agreement between ANN outputs and FEA data:

• Regression plots show R² values > 0.98, indicating near-perfect linear correlation.

• Error histograms display tightly clustered deviations near zero.

• Prediction errors remain below 3%, confirming reliability.

• ANN accurately captured nonlinear behaviors such as load-dependent stiffness variations.

This proves that the ANN can effectively replace repeated FEA simulations during design iterations.