Decision tree-based approach to extrapolate life cycle inventory data of manufacturing processes

Area of Science:

• Environmental Science
• Manufacturing Engineering
• Data Science

Background:

• Life cycle assessment (LCA) is vital for green manufacturing, but limited life cycle inventory (LCI) data hinders environmental burden analysis.
• Existing LCI databases often lack comprehensive data for specific manufacturing processes, posing a significant challenge.

Purpose of the Study:

• To develop and validate a novel machine learning approach for extrapolating LCI data, specifically greenhouse gas (GHG) emissions.
• To identify key factors influencing GHG emissions and resource consumption in manufacturing processes.

Main Methods:

• Utilized decision tree-based supervised machine learning models (Decision Tree, Random Forest, Gradient Boosting, Adaptive Boosting) trained on Ecoinvent LCI data.
• Employed correlation analysis to identify influential factors, data preprocessing, train-test splitting (70/30), and five-fold cross-validation for hyperparameter tuning.
• Evaluated models based on predictive performance using R-squared, Root Mean Squared Error, and Mean Percentage Error.

Main Results:

• The Gradient Boosting model demonstrated superior performance in extrapolating GHG emission data, achieving R-squared values <0.95 on the test set.
• Correlation analysis revealed workpiece material and manufacturing technology significantly impact resource consumption (energy, material, water).
• Energy consumption, water usage, and raw aluminum depletion were identified as key drivers of GHG emissions.

Conclusions:

• The proposed Gradient Boosting (GraBoost) model offers a robust computational method for estimating and extrapolating GHG emissions when LCI data is scarce or unavailable.
• This approach supports more accurate LCA and facilitates improved environmental management in green manufacturing.
• Identifying influential factors aids in targeted strategies for reducing environmental impact in manufacturing.