Author(s) :

Aadhav Hariish.

Volume/Issue :

Abstract :

Reducing unplanned equipment failures is critical for improving efficiency and lowering operational costs in manufacturing systems. This study investigates the use of machine learning models to predict machine failure using sensor-based operational data. The AI4I 2020 Predictive Maintenance Dataset was used, which includes temperature, rotational speed, torque, tool wear, and machine type variables. Exploratory data analysis identified key patterns, including nonlinear relationships between torque and failure probability and strong correlations among certain sensor variables. Four machine learning models, Logistic Regression, Random Forest, Gradient Boosting, and Support Vector Machine (SVM), were developed and evaluated. Due to class imbalance, performance was assessed using precision, recall, and F1-score, with emphasis on failure detection. While most models achieved high accuracy (0.97-0.99), Logistic Regression performed poorly in identifying failures (recall = 0.12). SVM achieved a high recall (0.84) but suffered from low precision (0.27), resulting in excessive false positives. A tuned Gradient Boosting model with threshold optimization achieved the best overall performance, with a precision of 0.77, a recall of 0.74, and an F1-score of 0.75. Receiver operating characteristic (ROC) analysis showed strong model discrimination, with an area under the curve (AUC) of 0.96 for Gradient Boosting and 0.95 for Random Forest. Compared to the untuned model, recall improved substantially, enabling the detection of more failure events with a moderate increase in false positives. These results demonstrate that ensemble methods, combined with threshold tuning, provide an effective approach for predictive maintenance by balancing failure detection and false alarm rates.