Aluminium ML Recipe Generator dataset (training results)

Dataset: Material Combinations and Chemical Properties

Overview

This dataset is generated by the ML Recipe Generator model in the Aluminium UC and represents different combinations of materials (recipes) and their resulting (generated) chemical composition, cost, and validity. The dataset is structured to help analyze the impact of different material mixes on final product properties and evaluate whether a given recipe meets the standarized norms (product criteria).

Columns Description

The dataset consists of the following key sections:

1. General Information

desired_quantity: Target quantity for production.

total_metal_input: Total amount of metal used in the recipe.

product_code: Identifier for the produced material (renamed for anonymization).

total_input: Total input materials used.

recalculated_yield: Adjusted yield after processing.

2. Material Composition

mat_1, mat_2, ..., mat_N: The proportions of different materials used in the recipe (renamed from original material names for anonymization).

3. Final Chemical Composition

These columns represent the final chemical composition of the produced material:

Final_Al (Aluminum)

Final_Fe (Iron)

Final_Si (Silicon)

Final_Mg (Magnesium)

Final_Mn (Manganese)

Final_Cu (Copper)

Final_Cr (Chromium)

Final_Pb (Lead)

Final_Ni (Nickel)

Final_Sn (Tin)

Final_Ti (Titanium)

Final_Zn (Zinc)

Final_Sb (Antimony)

Final_Ca (Calcium)

Final_Na (Sodium)

Final_P (Phosphorus)

Final_V (Vanadium)

Final_Zr (Zirconium)

Final_Sr (Strontium)

Final_Li (Lithium)

Final_B (Boron)

4. Cost Information

final_price: The total cost of the material combination.

5. Validity Flags

The dataset includes multiple validity checks performed by different models:

valid_general: General validity flag.

valid_gan: Validity check based on a the ML Recipe Generator (ctGAN) model.

valid_predicted: Predicted validity from the Chemical Analysis Predictor (Random Forest Regressor) model.

valid_theoretical: Theoretical validity based on predefined chemical rules i.e., mass-balance equations.

Usage

This dataset can be used for:

Analyzing material recipes and their feasibility.

Optimizing cost-effective material combinations.

Evaluating model-based predictions for material validity.

Training or validating machine learning models for material property predictions.

Notes

The material columns (mat_1 to mat_N) and material codes (prod_1 to prod_N) have been renamed from their original names to maintain consistency and confidentiality.

The dataset is structured to support various material science and manufacturing applications.

For any questions or additional clarifications, feel free to reach out!