Deep learning approach for energy efficiency prediction with signal monitoring reliability for a vinyl chloride monomer process. (March 2023)

• Record Type: Journal Article
• Authors: Panjapornpon, Chanin; Bardeeniz, Santi; Hussain, Mohamed Azlan
• Is Part Of: Reliability engineering & system safety. Volume 231(2023)
• Publication Date: 2023-03
• Subjects: Combined framework deep learning -- Fault detection and identification -- Energy efficiency prediction -- Petrochemical process -- Measurement reliability; Reliability (Engineering) -- Periodicals; System safety -- Periodicals; Industrial safety -- Periodicals; Fiabilité -- Périodiques; Sécurité des systèmes -- Périodiques; Sécurité du travail -- Périodiques; 620.00452
• DOI: 10.1016/j.ress.2022.109008
• Languages: English
• ISSNs: 0951-8320

Abstract

Highlights: A combined fault-detection and energy efficiency prediction model is developed. A VCM process in presence of faulty signal is considered as a case study. A fault-identifier signal enhances energy prediction accuracy and reliability. The model verifies by k-fold cross-validation and two benchmark datasets. The proposed model has robustness under excessive operating and fault variations. Abstract: A major concern for a real-time operation is the reliability of measurements. Especially for the petrochemical industry, which reveals complexity and uncertainty, the measurement fault causes consequences on safety, profitability, and utility management. Degraded signal quality not only leads to improper control action, but also creates more challenges for real-time energy efficiency management by reducing model performance and wasting more utility than standard operating practice. To improve system reliability and establish an effective energy efficiency monitoring tool, the combined framework for fault detection identification and energy efficiency prediction (FDI-EEP) based on a deep learning approach is proposed in this study. The FDI-EEP model uses the fault detection and identification result as a co-predictor for estimating energy efficiency aimed at improving the performance and reproducibility of the model and studying the effect of these faults on the downstream data-driven framework. Since process information is time-dependent, the long-short term memory layer is deployed on both networks to avoid gradient vanishing problems. A case study on the vinyl chloride monomer process datasets demonstrates that the proposed model precisely detected the measurement uncertainty and accurately performed the prediction task compared to other machine learning and prediction-based data cleaning methods. Graphical abstract: Image, graphical abstract