An AI-Assisted Framework for Intelligent Manufacturing Analytics and Decision Support Using ERP Data

Abstract

Manufacturing organizations record production, inventory, purchasing, and quality events in Enterprise Resource Planning (ERP) systems, yet these records are often fragmented, inconsistently timestamped, and reported with delay, which limits their usefulness for operational decisions. This paper proposes a modular framework that converts ERP transaction data into validated manufacturing key performance indicators (KPIs) and decision-support views, with an optional machine learning (ML) layer for production forecasting and anomaly detection. Following a design-science approach, the paper contributes (i) an eight-layer reference architecture covering ingestion, validation, an analytical data model, a KPI engine, optional ML, dashboards, and governance; (ii) an explicit catalogue of data-quality rules for missing values, duplicates, and inconsistent timestamps; (iii) a set of KPI definitions with stated variables, assumptions, and edge cases, aligned where possible with ISO 22400-2 and the availability-performance-quality decomposition of Overall Equipment Effectiveness; and (iv) a reproducible evaluation protocol based on a documented synthetic dataset, with naive and seasonal-naive forecasting baselines and labeled injected anomalies. The KPI formulas are verified against independently hand-calculated worked examples. The ML experiments are fully specified, and source code is provided, but they have not been executed for this version; no forecasting or anomaly-detection metrics are therefore reported. The framework has not been deployed or validated in a real manufacturing company. The paper discusses limitations, threats to validity, data governance, and privacy, and outlines the empirical work needed to validate the framework.