A Machine Learning-Based Decision Support System for Production Operations in Manufacturing Enterprises

Abstract

In manufacturing, real-time decision support for production operations is critical to optimize efficiency, minimize downtime, and ensure resource utilization. However, existing approaches fall short due to complex data integration and failure to model long-range temporal dependencies. To address these challenges, the authors propose the Transformer-based Decision Support Network (T-DSN), which integrates multi-source data and temporal modeling into a unified framework for predictive maintenance, resource allocation, and production scheduling. Leveraging the Transformer architecture, T-DSN captures long-term dependencies in time-series sensor data and operational logs, enabling accurate predictions and real-time decision-making. Experiments on SECOM, C-MAPSS, CMHS, and Purdue Production Scheduling datasets show T-DSN outperforms baselines like XGBoost and LSTM by up to 2% in R2 predictive accuracy with reduced training times, supporting efficient, cost-effective manufacturing operations.