Enhancing Product Quality in High-Variant Manufacturing: Combining Physics-Based Simulations and Data Science for...

Area of Science:

• Data Science
• Artificial Intelligence
• Physical Simulations
• Machine Learning

Background:

• Growing demands for quality, sustainability, and digitalization increase the importance of data science and AI across industries.
• Extensive product ranges present challenges, particularly for machine learning (ML) models with limited data per production variant.
• Existing approaches struggle to effectively analyze and model production variants with low quantities.

Purpose of the Study:

• To propose a methodology integrating data science and physical simulations for enhanced industrial data analysis.
• To enable machine learning (ML) and statistical analyses on process data across different production variants, especially those with limited data.
• To identify key parameters influencing product quality and explore their precise control using ML models.

Main Methods:

• Utilizing results from finite element method (FEM) simulations to transform process data.
• Developing a data transformation technique to ensure comparability of process data across diverse production variants.
• Applying the transformed data to machine learning (ML) methods and statistical analyses.

Main Results:

• Successfully demonstrated a method to effectively model production variants with very low quantities.
• Enabled the identification of critical parameters influencing product quality that are not evident through alternative methods.
• Illustrated the methodology's application using an aluminum production example.

Conclusions:

• The proposed methodology effectively overcomes data limitations for ML in industries with extensive product variations.
• This approach enhances production processes by identifying and enabling precise control over quality-influencing parameters.
• Significant potential exists for optimizing industrial processes through the integration of physical simulations and ML, despite some challenges.