Authors: Navya Kulshreshtha
Abstract: The rapid evolution of Industry 4.0 has necessitated a transition from traditional administrative Enterprise Resource Planning (ERP) to "Intelligent ERP" systems that leverage real-time operational data. This review article investigates the optimization of ERP performance through the integration of Machine Learning (ML)–based Predictive Maintenance (PdM) models. While traditional maintenance strategies within ERP namely reactive and preventive often lead to unplanned downtime or resource wastage, ML-based PdM offers a data-driven alternative that predicts equipment failure before it occurs. This study synthesizes current literature regarding the architectural integration of Industrial Internet of Things (IIoT) sensors with ERP modules, such as Asset Management, Production Planning, and Materials Management. We categorize the predominant ML methodologies including Supervised Learning for fault classification, Deep Learning (LSTM and GRU) for Remaining Useful Life (RUL) estimation, and Unsupervised Anomaly Detection evaluating their specific contributions to enterprise-level efficiency. The review highlights how PdM-driven insights directly optimize ERP Key Performance Indicators (KPIs) by reducing maintenance costs, streamlining spare parts inventory through Just-in-Time (JIT) procurement, and enhancing Overall Equipment Effectiveness (OEE). Furthermore, the article addresses critical implementation challenges, such as data silos, scalability, and the "black box" nature of AI models. By analyzing the synergy between predictive analytics and resource orchestration, this review provides a roadmap for researchers and practitioners to build resilient, self-optimizing industrial ecosystems. The findings suggest that the integration of ML-PdM is no longer a peripheral technical upgrade but a core strategic necessity for modern enterprise resource management, enabling a shift from descriptive reporting to prescriptive action.
DOI: http://doi.org/10.5281/zenodo.18159637
