Digital Twin–Enabled Pressure Optimization for Energy, Cost, and Carbon Reduction in Centrifugal Compressors: Industrial Validation in Semiconductor Manufacturing

ABSTRACT

Compressed dry air (CDA) is one of the most energy‐intensive utility systems in semiconductor manufacturing because it operates continuously to satisfy stringent production and cleanroom requirements. In many facilities, centrifugal compressors are intentionally operated at discharge pressures above the actual process demand to ensure a reliable air supply. Although this practice enhances operational stability, it also increases electricity consumption, operating costs, and associated carbon emissions. This study proposes and validates a hybrid digital twin framework for improving compressor energy performance through discharge pressure optimization. The framework integrates thermodynamic modelling, real‐time operational data, and predictive analytics to evaluate compressor behaviour under different operating conditions. A dataset comprising 20,577 operational records collected from a semiconductor manufacturing facility in Malaysia was used for industrial validation. During the investigation, the compressor discharge pressure was progressively reduced from 8.9 bar to 7.5 bar while maintaining production demand and airflow requirements. Under these conditions, electricity consumption decreased by 10.25%, whereas the delivered airflow remained stable, indicating that the original operating pressure provided an unnecessary pressure margin. Statistical evaluation using multivariate analysis of variance (MANOVA) together with regression analysis identified discharge pressure as the dominant factor influencing energy consumption, while airflow variations were not statistically significant. The digital twin achieved a prediction error of less than 5%, demonstrating close agreement between simulated and measured operating performance. In addition to improving energy efficiency, the optimized operating strategy reduced operating costs and lowered indirect carbon dioxide (CO ₂ ) emissions by approximately 10%. Unlike many earlier studies that relied primarily on simulation or laboratory‐scale investigations, this work demonstrates the practical application of a digital twin using real operational data from a full‐scale semiconductor manufacturing environment. The proposed framework offers an effective decision‐support approach for enhancing energy efficiency, reducing operating costs, and advancing sustainability objectives in semiconductor manufacturing.