Integrating Dependency Analysis Through Structural Equation Modeling and Artificial Neural Networks: A Case Study in the Mining Industry

Industry 4.0 technologies are revolutionizing industrial maintenance management, highlighting Machine Learning (ML) techniques as key tools to anticipate failures more efficiently. In this study, the dependencies between components of a crushing line of a mining company in Chile and the different types of failures are analyzed, using ML models and structural equation models (SEM), with the objective of determining which ML model best fits the data, providing reliable relationships, so that in future work these relationships can be used in failure prediction models. Both models complement each other, since it is currently recognized the importance of a comprehensive approach in the analysis of failure types, allowing to improve maintenance management by offering an alternative to reduce costs associated with maintenance and downtime. The main motivation is to increase the accuracy of early warning systems, supporting more informed decision making. ML models such as Random Forest (RF) and Artificial Neural Networks (ANN) are employed, whose dependency analyses have shown positive results in previous studies. In addition, Structural Equation Modeling (SEM) is integrated, which allows exploring the complex interrelationships between system variables and different types of faults. The models were evaluated using the confusion matrix, accuracy, precision, recall and F1 score, complemented by SEM-derived indicators that reinforce the validity of the results. ANN showed outstanding performance with an accuracy of 0.9926 and significant relationships according to SEM, whereas RF suffered from overfitting, limiting its applicability in SEM. This dependence analysis provides a novel approach, using two techniques that together provide a more robust analysis of dependence research and contributing to existing research in this field.

Keywords: Machine learning, Structural equation model, Dependency analysis.