Abstract
Industrial production systems require intelligent maintenance solutions to minimize unplanned downtime, improve reliability, and support data-driven decision making. This study presents a machine learning framework for condition monitoring and fault classification in industrial production systems. The framework is evaluated using an AI4I-derived synthetic dataset comprising 10,000 production events characterized by operational sensor measurements and machine failure indicators. An exploratory analysis is first conducted to examine data distributions, failure patterns, and relationships among operational variables. Subsequently, four classification models (Logistic Regression, Random Forest, Histogram-Based Gradient Boosting, and a Multilayer Perceptron (MLP) neural network) are developed and comparatively evaluated. The results show that nonlinear models significantly outperform the Logistic Regression baseline, with Random Forest, Histogram-Based Gradient Boosting, and MLP achieving very high classification performance. The findings suggest that interactions among operational variables contribute substantially to the fault classification task and are more effectively captured by nonlinear learning approaches than by linear models using the original feature set. Overall, the study provides a benchmark-style comparative evaluation of representative machine learning classifiers for fault classification on an AI4I-derived synthetic dataset. The findings primarily illustrate classifier behavior under controlled synthetic conditions and provide a basis for future validation using real industrial data and operational maintenance environments.