Optimization of the Fabrication of Cold Drawn Steel Wire Through Classification and Clustering Machine Learning Algorithms

The demanding deformations steel is subjected to during drawing may result in the breakage of the wire. The hypothesis of this research is that drawing failure is not a random event but can be predicted using a suitable approach. Machine Learning classification and clustering algorithms have been implemented to predict the probability of failure during drawing and to optimize the manufacturing conditions to reduce the failure rate. The following algorithms have been employed for classification: K-Nearest Neighbors, Random Forests and Artificial Neural Networks. The reduced value of the rejection rate implies that classification must be carried out on an imbalanced dataset. For this reason, resampling methods (undersampling, oversampling and SMOTE) and specific scores for imbalanced datasets were used. It was possible to obtain a qualified Random Forest classifier which provided satisfactory scores (ROC AUC of 0.824 and an average precision of 0.604 in the test dataset). This tool allows the heats with a higher probability of undergoing any breakage during drawing to be detected, thus improving the final quality of the product. K-means clustering (K = 4) has been successfully used in this study to identify those manufacturing conditions that minimize the number of breakages during drawing. The results of the clustering analysis show that the rate of heats undergoing failure may be reduced by a factor of 2.5.