Using an artificial neural network to predict the residual stress induced by laser shock processing

Area of Science:

• Materials Science and Engineering
• Mechanical Engineering
• Computational Modeling

Background:

• Laser shock processing (LSP) is a surface treatment technique used to enhance material properties.
• Understanding and predicting residual stresses induced by LSP is crucial for material performance.
• GH4169, a Ni-Cr-Fe-based precipitation-hardening superalloy, is widely used in demanding applications.

Purpose of the Study:

• To develop and validate an artificial neural network (ANN) model for predicting residual stresses in GH4169 superalloy.
• To investigate the influence of LSP parameters (power density, overlap rate) on induced residual stresses.
• To assess the accuracy of the ANN model in predicting depth-wise residual stresses.

Main Methods:

• Experimental samples of GH4169 superalloy were subjected to LSP with varying laser power densities and overlap rates.
• Depth-wise residual stresses were measured using the x-ray diffraction sin²ψ method combined with layer-by-layer electrolytic polishing.
• An ANN model was constructed with laser power density, overlap rate, and depth as inputs, and residual stress as the output.

Main Results:

• LSP successfully introduced beneficial compressive residual stresses in the near-surface region of GH4169.
• Surface residual stresses ranged from -236 MPa to -799 MPa, with compressive stress depths exceeding 0.50 mm.
• The ANN model demonstrated high prediction accuracy, with R² values of 0.9948 for training sets and 0.9931 for testing sets.

Conclusions:

• The ANN method is highly accurate for predicting residual stresses induced by LSP in metallic materials.
• The developed model provides valuable insights for optimizing LSP process parameters.
• This approach can guide the enhancement of material properties through controlled residual stress introduction.