Peen-ML: Using Machine Learning to Replace Dynamic Simulation for Predicting Shot Peening Deformation

Peen-ML is a solution that uses machine learning to predict shot peening process deformation. By training on simulation data and empirical datasets, it provides a faster alternative to traditional dynamic simulation, aiming to solve the problems of high computational cost and long time consumption in traditional simulation. The project is maintained by onestr1, published on GitHub, and licensed under MIT.

Shot peening is widely used in aerospace, automotive manufacturing, and other fields, which can improve material fatigue life and stress corrosion resistance. However, traditional dynamic simulation has problems such as long computation time (several hours to days per run), high resource consumption (requiring high-performance clusters), and difficult iteration (high cost of re-simulation after parameter adjustment), which restrict the efficiency of process parameter optimization.

The core idea of Peen-ML is to replace traditional dynamic simulation with machine learning models: by training on a large amount of simulation data and empirical datasets, it learns the mapping relationship between shot peening process parameters and deformation. After the model is trained, predictions can be completed in milliseconds, achieving an order-of-magnitude speedup compared to traditional simulation. Training data sources include simulated shot peening effect data, residual strain models, and empirical datasets.

The project is implemented in Python, with a code size of approximately 83KB. The model architecture is based on data-driven supervised learning methods to learn the nonlinear mapping from input process parameters to output deformation; data processing integrates multi-source data (simulation results and measured empirical data) to build a unified training set; the prediction process is: input process parameters → model inference → output deformation prediction results.

The value of this method is reflected in: 1. Process parameter optimization: quickly evaluate the effect of different parameter combinations and find the optimal window; 2. Quality control: predict deformation before production, adjust parameters in advance to avoid out-of-tolerance; 3. Cost reduction: reduce simulation computing resource consumption; 4. Real-time decision-making: support online process adjustment to meet the needs of on-site rapid decision-making.

Limitations: 1. Data dependence: accuracy is highly dependent on the quality and coverage of training data; 2. Generalization ability: predictions for extreme process parameters may have deviations; 3. Physical consistency: the model may only learn statistical correlations and may not strictly satisfy physical laws. Future directions: neural networks combined with physical constraints, active learning strategies to reduce the need for labeled data, and uncertainty quantification to evaluate prediction reliability.

Peen-ML represents an important trend in the field of engineering simulation: using data-driven methods to supplement or replace traditional physical simulation. This hybrid paradigm of "simulation + machine learning" is expected to greatly improve computational efficiency while maintaining accuracy, bringing new possibilities to engineering design.