VSG-FC: A Combined Virtual Sample Generation and Feature Construction Model for Effective Prediction of Surface...

Area of Science:

• Manufacturing Engineering
• Materials Science
• Artificial Intelligence

Background:

• Surface roughness is crucial for workpiece quality and performance.
• Accurate prediction enhances production efficiency.
• Data-driven models for surface roughness prediction typically require extensive training data, which is often difficult to obtain in polishing processes due to cost and time constraints.

Purpose of the Study:

• To propose a novel data-driven model, VSG-FC, for predicting surface roughness.
• To address the challenge of data scarcity in polishing processes.
• To enhance prediction accuracy and identify key machining factors through optimized feature spaces.

Main Methods:

• The study integrates Genetic Algorithm-driven Virtual Sample Generation (GA-VSG) for sample augmentation.
• It also incorporates Genetic Programming-driven Feature Construction (GP-FC) for feature reconstruction.
• The combined VSG-FC approach optimizes the feature space to improve model performance.

Main Results:

• The VSG-FC method demonstrated significant advantages in generating high-quality virtual samples.
• The model achieved enhanced prediction accuracy for surface roughness.
• The approach proved to be explainable, successfully identifying critical machining factors influencing surface roughness.

Conclusions:

• The proposed VSG-FC model effectively overcomes data scarcity in surface roughness prediction.
• The integration of GA-VSG and GP-FC optimizes feature spaces for improved model performance.
• The method offers a viable, explainable solution for enhancing production efficiency and product quality in polishing processes.