Inverse solution of process parameters in gear grinding using hierarchical bayesian physics informed neural network...

Area of Science:

• Manufacturing Engineering
• Computational Science
• Artificial Intelligence

Background:

• Accurate inverse solutions for process parameters are vital in precision gear grinding.
• Model parameter correlations and uncertainties complicate inverse solutions.
• Existing methods struggle with the complexity of gear grinding parameter prediction.

Purpose of the Study:

• To propose a novel Hierarchical Bayesian Physics-Informed Neural Network (HBPINN) for inverse solutions in gear grinding.
• To address challenges of complex correlations and uncertainties in model parameters.
• To improve the accuracy and efficiency of predicting process parameters from surface roughness data.

Main Methods:

• Developed a global-group-individual hierarchical structure for model parameters.
• Employed a hierarchical Bayesian framework for correlation and uncertainty analysis.
• Integrated multivariate regression and Kullback-Leibler divergence into a physics loss function.
• Utilized Gaussian Process Regression (GPR) for dataset generation.

Main Results:

• HBPINN demonstrated superior efficiency and accuracy compared to BPINN, VI-BPINN, and PINN.
• Achieved an average R² of 0.9629 with a training set size of 200.
• Reduced prediction time by 4-10 times.
• Exhibited excellent uncertainty quantification and robustness.

Conclusions:

• HBPINN effectively solves inverse problems in precision gear grinding.
• The proposed hierarchical structure and physics-informed approach enhance predictive capabilities.
• HBPINN offers a robust and efficient solution for complex manufacturing process optimization.