Machine Learning Tool for Predicting Defect Occurrence from Mechanical Test Data

# Machine Learning Tool for Predicting Defect Occurrence from Mechanical Test Data ## Description This repository contains a dataset and a machine learning-based predictive tool for assessing the occurrence of defects in metallic materials based on mechanical test data. The predictive model was developed using supervised machine learning techniques and is intended to support decision-making in material selection and forming process design. Predictions are based on mechanical properties obtained from: - Tensile tests performed along the rolling direction and transverse direction- Hole expansion tests The model provides: - Binary classification (defect / no defect)- Probability of defect occurrence --- ## Machine Learning Model The predictive framework is based on an ensemble learning strategy using soft voting across multiple classifiers implemented with the **scikit-learn** library. The ensemble combines the following models: - Multi-layer Perceptron (MLP)- Support Vector Machine (SVM)- Random Forest- Logistic Regression- Gaussian Naive Bayes- Decision Tree- K-Nearest Neighbours The final prediction corresponds to the probabilistic consensus of these classifiers. --- ### Files | File | Description ||-----|-------------|| Dataset.xlsx / Dataset.csv | Dataset used to train the machine learning model || test.xlsx / test.csv | Input file containing properties of materials to be evaluated || Prediction.exe | Executable program that performs the prediction || Results.xlsx | Output file containing predicted class and probability || User_Guide.pdf | Instructions for using the prediction tool | --- ## Workflow 1. Define the defect criterion in the dataset (deformation threshold).2. Add new materials to the dataset if required.3. Save the dataset as `Dataset.csv`.4. Insert the properties of the material to evaluate in `test.xlsx`.5. Save the file as `test.csv`.6. Run `Prediction.exe`.7. Predicted results are written to `Results.xlsx`. --- ## Output The prediction tool returns: | Output | Description ||------|-------------|| Prediction | 0 = no defect / 1 = defect || Probability | Estimated probability of defect occurrence | --- ## Project Context This dataset and predictive tool were developed within the SafeForming project: SafeForming – Intelligent System for Defect Prevention in Cold Stamping Components Reference: POCI-01-0247-FEDER-017762 The project aimed to develop intelligent methodologies capable of predicting the occurrence of defects (such as cracking) in cold stamped sheet metal components through the integration of material characterization, process parameters and machine learning techniques. Project website:https://safeformingproject.wordpress.com/--- ## Related Publications The dataset and predictive framework are related to the following publications on machine learning applications in sheet metal forming: Contente, J.; Prates, P. Predicting edge cracking in sheet metal forming: evaluating machine learning models and data transformations. International Journal of Advanced Manufacturing Technology, 2025. https://doi.org/10.1007/s00170-025-15721-6 Marques, A. E.; Prates, P.; Fonseca, A. R.; Oliveira, M. C.; Soares, M. S.; Fernandes, J. V.; Ribeiro, B. Machine Learning for Predicting Fracture Strain in Sheet Metal Forming. Metals, 12(11), 1799, 2022. https://doi.org/10.3390/met12111799 Marques, A. E.; Prates, P.; Fonseca, A. R.; Oliveira, M. C.; Soares, M. S.; Fernandes, J. V.; Ribeiro, B. Machine Learning for the Prediction of Edge Cracking in Sheet Metal Forming Processes. In: Machine Learning and Artificial Intelligence with Industrial Applications. Springer, 2022. https://doi.org/10.1007/978-3-030-91006-8_6 --- ## Authors Pedro Prates (Univ. Aveiro); Armando Marques (Univ. Coimbra); Marta Oliveira (Univ. Coimbra); Bernardete Ribeiro (Univ. Coimbra); José Valdemar Fernandes (Univ. Coimbra); --- ## License This dataset and software are made available for academic and research purposes. Please cite the repository if it contributes to your work.