Neural Network-Enabled Process Flowsheet for Industrial Shot Peening.

Area of Science:

• Materials Science
• Mechanical Engineering
• Computational Modeling

Background:

• Shot peening is crucial for inducing residual stress in components.
• The evolution of peening media (size, shape) dynamically affects stress distribution.
• Spatially heterogeneous residual stress fields result from stochastic particle impacts.

Purpose of the Study:

• To develop a dynamic flowsheet model for predicting residual stress in shot peening.
• To enable real-time, mechanistically grounded predictions of surface stress.
• To investigate the impact of media recharge strategies on residual stress outcomes.

Main Methods:

• A dynamic flowsheet model was developed to simulate peening media evolution.
• A convolutional long short-term memory (ConvLSTM) neural network was trained on finite element simulations.
• The model was validated in industrial 10,000-cycle production peening case studies.

Main Results:

• The model accurately predicts spatially heterogeneous residual stress fields.
• Real-time prediction of surface stress evolution under industrial conditions was achieved.
• Media recharge strategy was demonstrated to have a measurable effect on residual stress outcomes.

Conclusions:

• Dynamic modeling and neural networks offer a powerful approach for predicting shot peening residual stress.
• Optimizing media recharge strategies can control and enhance residual stress distribution.
• This work provides a framework for mechanistically grounded, fast prediction of surface stress evolution.